full cmake 7/N: library

[loriab]

currently atop #315 . new base lib in ee93781

• [x] build the export tarball. uses either gzip or bzip2 (closes #316 ). largely imitates the libtool tarball -- see differences below, mainly in positioning of fortran files

>>> diff ../libint2-upstream-wip/tarball_contents_281_full tarball_contents_current10 | grep ">"
> libint-2.8.1/include/libint2/config2.h.cmake.in
> libint-2.8.1/include/libint2/config.h
> libint-2.8.1/cmake/modules/FindMultiprecision.cmake
> libint-2.8.1/cmake/modules/autocmake_safeguards.cmake
> libint-2.8.1/cmake/modules/int_orderings.cmake
> libint-2.8.1/cmake/modules/int_computed.cmake
> libint-2.8.1/cmake/modules/FindEigen3.cmake
> libint-2.8.1/cmake/modules/int_userreal.cmake
> libint-2.8.1/cmake/modules/options.cmake
> libint-2.8.1/cmake/modules/int_checkboost.cmake
> libint-2.8.1/features
> libint-2.8.1/tests/CMakeLists.txt
> libint-2.8.1/tests/fortran/
> libint-2.8.1/tests/fortran/test-eri.cc
> libint-2.8.1/tests/fortran/test.cc
> libint-2.8.1/tests/fortran/fortran_example.F90
> libint-2.8.1/tests/hartree-fock/hftest.cmake
>>> diff ../libint2-upstream-wip/tarball_contents_281_full tarball_contents_current10 | grep "<"
< libint-2.8.1/include/libint2/config.h.in
< libint-2.8.1/include/CMakeLists.txt
< libint-2.8.1/compiler.config
< libint-2.8.1/cmake/hftest.cmake
< libint-2.8.1/cmake/modules/RedefaultableOption.cmake
< libint-2.8.1/cmake/modules/FindMPFR.cmake
< libint-2.8.1/cmake/modules/JoinPaths.cmake
< libint-2.8.1/cmake/libint2.pc.cmake.in
< libint-2.8.1/MakeVars.features
< libint-2.8.1/fortran/test-eri.cc
< libint-2.8.1/fortran/test.cc
< libint-2.8.1/fortran/fortran_example.F90

• [x] GHA run with a variety of shared/static, release/debug, subproject=on/off . steps are broken up more to detect failures earlier, and tests are run with a bit of parallelism. The export job of GHA is reenabled.
• [x] build base library (shared & static) and header-only cxx intf (linked to shared & static). Controlled through BUILD_SHARED_LIBS and LIBINT2_BUILD_SHARED_AND_STATIC_LIBS for both.
  • [x] generating ints into a new dir generated/ rather than directly in src/ so that not removing headers from the dir once generated. I suspect that since all lib builds going through the tarball that __COMPILING_LIBINT could be removed after editing the export structure.
  • [x] all targets assembled in order

• target_compile_features(
• target_compile_definitions(
• target_compile_options(
• set_target_properties( # note cmake generator expressions not allowed for this cmd
• target_include_directories(
• target_link_libraries(

• [x] enable fortran
• [x] enable python
  • [x] fixed #319
  • [x] linking to header-only cxx rather than to base object library + eigen
  • [x] added a linux-style install-to-prefix for packagers
  • [x] python module not built by default even if python enabled
• [x] enable compiled cxx library
  • [x] I finally succeeded in making the compiled cxx lib link to something, namely tests!
  • [x] lib unit tests and the hartree-fock (plain) and the fortran cxx tests currently use compiled when available and header-only when not, rather than duplicating the whole test suite.
  • [x] I had to add in an extra define for the tests to allow it to compile. Is this right in general?
  • [x] linking the compiled cxx to the hartree-fock++ test on my local Linux box (conda gcc12) always works, but on GHA, it succeeds with Linux+conda-gcc and fails with Linux+ubuntu-gcc10 or mac or windows or intel). See #239
  • [x] the compiled library had only ever been in #233, never released. It is controlled by the following, which I'm toggling to off by default LIBINT2_REQUIRE_CXX_API_COMPILED=OFF
• [x] all the optional what-to-build features standardized to LIBINT2_ENABLE_ prefixes since ENABLE_PYTHON is pretty generic and prone to conflicts with FetchContent.
• [x] Updated DynamicVersion from upstream ... see a choice at https://github.com/LecrisUT/CMakeExtraUtils/issues/24#issuecomment-1892851236
• [x] changed some // comments in ifdefs to /* ... */ since the former were raising warnings with some compilers
• [ ] stuff that doesn't work
  • [ ] mac dylib with fortran #318
  • [ ] mpfr -- I haven't gotten the lib unit tests to compile with LIBINT_HAS_MPFR=1
  • [ ] c++ compiled lib for hartree-fock++ (see above)